Data-driven and privacy-preserving risk assessment method based on federated learning for smart grids.

Area of Science:

• Electrical Engineering
• Computer Science
• Cybersecurity

Background:

• Smart grids generate vast high-dimensional data, challenging conventional risk assessment methods.
• Centralized risk evaluation in smart grids raises privacy concerns and operator reluctance to share data.
• Existing methods struggle with data volume and privacy preservation for security risk evaluation.

Purpose of the Study:

• To develop a data-driven, privacy-preserving method for smart grid security risk assessment.
• To address limitations of conventional methods in handling large datasets and protecting sensitive information.
• To enhance operational planning and threat detection in smart grids through accurate, private risk evaluation.

Main Methods:

• Implemented a two-tier risk indicator system and an expanded dataset for comprehensive analysis.
• Utilized a deep convolutional neural network (CNN) model to analyze system variables and risk levels.
• Developed a secure federated risk assessment protocol incorporating homomorphic encryption for parameter protection.

Main Results:

• The proposed method demonstrated high accuracy in security risk assessment on IEEE 14-bus and IEEE 118-bus systems.
• Experimental results confirmed the effectiveness of the federated learning approach in safeguarding data privacy.
• The integration of deep learning and secure encryption successfully protected model parameters during training.

Conclusions:

• The novel federated learning framework effectively balances accurate smart grid risk assessment with robust data privacy.
• This approach overcomes challenges posed by large datasets and privacy concerns in conventional methods.
• The method offers a reliable solution for secure operational planning and threat detection in smart grids.